SAN FRANCISCO — OpenAI has temporarily paused some internal activities involving its upcoming Astra artificial intelligence model after preliminary cybersecurity evaluations indicated that the system may have capabilities approaching the company's highest-risk cyber threshold.

OpenAI said the early evaluations showed Astra performing strongly enough in cybersecurity that the company could not rule out a "Critical" capability level under its Preparedness Framework. The company stressed that Astra is still an upcoming model and that the evaluations remain preliminary.

The development has placed AI cybersecurity and model safety at the center of attention as increasingly capable AI systems become better at identifying vulnerabilities and performing sophisticated technical tasks.

OpenAI said it is introducing stronger security controls for higher-capability models. These include isolated testing environments, restricted network and tool access, enhanced protection of model weights, additional monitoring and sandboxed execution.

The company has also implemented monitoring for risky actions and potential misalignment across Astra's agentic applications, including training and evaluation activities. According to OpenAI, these systems are designed to identify high-risk behavior and trigger security responses when necessary.

The announcement highlights a growing challenge for the U.S. technology industry: AI systems can become increasingly useful for cybersecurity defenders while simultaneously creating new risks if advanced capabilities are misused.

OpenAI said it plans to work with relevant government agencies and selected AI safety organizations to test Astra's capabilities. The company also plans to provide recommended security controls to third-party testing partners handling higher-risk evaluations.

The move comes as technology companies race to develop more capable AI agents that can perform complex tasks with greater independence. As these systems become more powerful, cybersecurity safeguards are becoming an increasingly important part of the development process.

For the U.S. technology sector, the Astra announcement could become an important example of how frontier AI companies manage the transition from experimental capabilities to responsible deployment.

The immediate focus is no longer simply how powerful the next AI model can become, but whether its security infrastructure can keep pace with that power.